AI summary by TwinMind
Cosine Similarity, Clearly Explained!!!
StatQuest with Josh Starmer
10m 13s
Introduction & Context
- Josh Starmer presenting StatQuest educational video on cosine similarity
- Sponsored by letters A, B, C (Always Be Curious motto)
- Topic: cosine similarity metric for measuring text/document similarity
Problem Setup & Motivation
- Example: analyzing sentences about 1990 movie Troll 2
- First 3 sentences express positive feelings, last sentence negative
- Easy to identify similarities manually with few sentences
- Challenge: analyzing large datasets (e.g., month of Twitter traffic)
- Need automated computer-based similarity measurement
Basic Cosine Similarity Concept
- Cosine similarity: relatively easy metric to calculate similarity/difference
- Measures angle between vectors in multi-dimensional space
- Values range from 0 (completely different) to 1 (identical)
Simple Two-Word Example
- Compare phrases: "hello world" vs "hello"
- Step 1: Create word count table
- "hello world": hello=1, world=1
- "hello": hello=1, world=0
- Step 2: Plot on 2D graph (hello on x-axis, world on y-axis)
- "hello world" point at (1,1)
- "hello" point at (1,0)
- Step 3: Draw lines from origin (0,0) to each point
- Step 4: Calculate angle between lines = 45 degrees
- Cosine similarity = cos(45°) = 0.71
Key Properties of Cosine Similarity
- Angle-dependent, not length-dependent
- Example: "hello hello hello" vs "hello world"
- Point moves to (3,0) but angle remains 45°
- Cosine similarity still = 0.71
- Identical phrases: angle = 0°, cosine similarity = 1
- No common words: angle = 90°, cosine similarity = 0
- Partial overlap: cosine similarity between 0 and 1
Mathematical Formula
- Direct calculation formula provided (complex sigma notation)
- Variables explained:
- i = index (word position), starts at 1
- n = total number of different words
- A, B = the two phrases being compared
- Example calculation yields 0.71 (matches geometric approach)
Multi-Dimensional Extension
- Two words = 2D graph (feasible to visualize)
- More words = higher dimensions (impossible to visualize)
- Example: "I love Troll 2" vs "I love Gym Kata"
- 5 different words = 5-dimensional space
- Formula still works: cosine similarity = 0.58
- Formula advantage: works regardless of dimensionality
Movie Reference
- Gym Kata: 1985 movie about gymnast using parallel bars in Mortal Kombat
Promotional Content
- StatQuest PDF study guides available
- Book: "StatQuest Illustrated Guide to Machine Learning"
- Available at statquest.org
- Support options: Patreon, channel membership, merchandise, songs, donations
